In this work, two methods are adopted and compared to learn a helicopter hover controller subjected to control input delay and noise in state estimate: neural network hover controller described in [Ng, Kim, Jordon, & Sastry, 2003], and the cross entropy algorithm described in previous work.
The controller was capable to work with noise and latency to certain manner. The best performance was shown below:
Shih-Yun (Laila) Lo 